Trade and Communication Under Subjective Information
Jack Douglas Stecher1
January 2006
1Norwegian School of Economics and Business Administration, Department of Accounting, Auditing and
Law, Helleveien 30, N-5045 Bergen, Norway. Email: [email protected]
Abstract
This paper models an economy where agents perceive the choices they face subjectively, and have
subjective interpretations of the terminology they use in a shared business language. Preferences
are defined on what an agent perceives, and not on what is objectively presented to an agent.
A business language enables agents to trade, provided the terminology in the language is sufficiently
vague: once agents can express more detail than their trading partners can perceive, the language
ceases to be useful. Under some regularity conditions on the language, an appropriately defined
notion of competitive equilibrium exists. However, much less can be said about welfare than in the
neoclassical case, as there are counter-examples to both welfare theorems.
Key Terms: Language, Perceptual Limits, Unawareness
JEL Classifications: D820, M410, D010, C650
1
1 Introduction
For economic activity to occur, both parties in a trade need to have a shared understanding of what
is being exchanged. The economic theorist’s usual starting point is assuming that the commodity
space is common knowledge and that agents have preference relations defined directly on this
commodity space. While agents may fail to consider alternatives, as in the literature on unawareness
(Fagin and Halpern 1988, Dekel, Lipman, and Rustichini 1998a, Modica and Rustichini 1999), they
must still have preferences defined over commodity bundles described in arbitrarily fine detail. That
is, agents must know in principle everything that could become available to them, in any state of
the world, at any location or date.
In such a world, agents in principle can describe everything they might offer or desire in trade.
There may be strategic reasons that reliable communication is restricted (Crawford and Sobel
1982), social benefits of restricted communication (Kanodia, Singh, and Spero 2005), or technical
restrictions placed on the language (Rubinstein 2000), but there is no inherent reason why an agent
cannot know infinitely fine details about what could be traded.
It is easy to imagine situations where subtle distinctions among possible choices are lost on an
agent, and in fact where the same agent may be able to perceive the same object in multiple ways.
For example, an agent might be unable to taste the difference between a medium-bodied wine and a
light-bodied wine, if the two wines differ relatively little in their sugar and glycerin concentrations.
On the other hand, the same agent may be unable to taste the difference between the same medium-
bodied wine and a full-bodied wine, again for sufficiently small differences in the concentrations of
these substances. Thus a fixed bottle of wine, which objectively may be medium-bodied, can be
perceived as both a full-bodied wine and a light-bodied wine. Whether the agent knows the wine’s
objective sugar and glycerin concentrations is not what necessarily determines how well the agent
enjoys the wine; instead, how the agent subjectively perceives the wine’s body is what matters.
Conversely, an agent facing distinct real-world objects may perceive them in the same way: in
the above example, both the full-bodied wine and the light-bodied wine taste the same, at least
sometimes. These two phenomena underlie the notion of a just noticeable difference. It has been
observed that non-degenerate just noticeable differences are problematic for the assumption of
2 2 PERCEPTION
complete, transitive preferences on the objective commodity space (Mas-Colell, Whinston, and
Green 1995, Knoblauch 1998, Dubra, Maccheroni, and Ok 2004).
As different agents can have different perceptions, a business language will typically enable agents to
describe their proposed trades only approximately. The same perceived object might be reportable
in multiple ways—a given bottle of wine might justifiably be called full-bodied or medium bodied—
and conversely the same description might apply to multiple objects. These two phenomena are the
basis of the accountant’s notion of an immaterial difference. Unless everyone perceives the world
in the same way, it turns out to be useful for a business language to have a non-trivial notion of
immaterial differences: vague requests can be fulfilled more reliably than excessively detailed ones.
In the above wine example, imagine that a connoisseur asks the advice of a novice wine steward.
The connoisseur will discover that an overly precise request is one that the wine steward cannot
reliably fill. After enough unpleasant surprises, the connoisseur presumably will learn that it would
be better to requests that the steward cannot get wrong.
This paper formalizes these ideas, in order to study the economic consequences of subjective infor-
mation. The next section presents an analytic model of an agent’s subjective perceptions. Section
three models the use of a shared business language among agents with private, subjective views
of the world, and provides conditions under which agents can receive in trade what they believe
they agreed upon. Section provides conditions for existence of an appropriately defined notion of
equilibrium, and presents counter-examples to both welfare theorems. Section five concludes.
2 Perception
2.1 Basic Model of Perception
Throughout this paper, there is assumed to be a set I of agents. The model of individual perception
presented here consists of two collections and a binary relation between them. The first collection
consists of possible consumption choices (“real-world objects”) labeled X. Agents do not observe
X, and in fact need not be aware of X; one can think of X as chiefly being in the model for the
researcher’s convenience.
2.1 Basic Model of Perception 3
For agent i ∈ I, there is a collection Si, known only to i, called i′s set of subjective conceptions.
When no confusion can arise, I drop the subscript and write S for an arbitrary agent’s subjective
conceptions. Agent i’s preferences are defined on Si, i.e., on the world as the agent understands it.
A perception is some conception that the agent observes when facing a real-world object. That
is, when an agent sees some x ∈ X, the agent perceives some subjectively meaningful a ∈ S. A
given x ∈ X need not have a unique conception a ∈ S as the only way it can be perceived, but
an agent is assumed to have coarse enough conceptions to enable every object to be perceived as
something. Conversely, not every conception a ∈ S is necessarily the perception of a unique x ∈ X.
In terms of the wine example from the introduction, there are individual bottles of wine that the
agent might in some contexts perceive as full-bodied and in others as light-bodied, and there are
individual conceptions of wine such as “full-bodied” that many wines may be perceived as.
Because of this many-to-many relationship, perception is modeled as a binary relation, written as
, between X and S.1 For x ∈ X and a ∈ S, the relation x a is read as, “real-world object x
can be perceived as subjective conception a.” With some technical assumptions, it is shown that
the agent’s perceptions induce a topology on X, with S as a base of this topology. Intuitively,
this means that the agent’s preferences are defined on approximate consumption bundles, where
the approximation is determined by the fineness of the agent’s conceptual framework S and by
the fineness of the agent’s perceptual apparatus . Thus the framework here is related to the
approximate price sensitivity in Allen and Thisse (1992), the model of consumer choice over sets in
Kreps (1979), and the approximate meanings of numerical stimuli in Dickhaut and Eggleton (1975)
There are two correspondences associated with the perception relation:
Definition 2.1. The correspondence X −→ S is given by
(∀x ∈ X) (x) ≡ {a ∈ S |x a}.
The inverse correspondence S −1
−→ X is given by
(∀a ∈ S) −1 (a) ≡ {x ∈ X |x a}.
1This symbol is the forcing relation introduced by Cohen. For background and historical discussion, see Avigard
(2004).
4 2 PERCEPTION
There are two possible notions of the image a subset D ⊆ X of the commodity space. One can
take the strong image, giving the members of S that are in the image of every point in D, or one
can take the weak image, giving the members of S that are in the image of some point in D.
Definition 2.2. The strong (or universal) image of D ⊆ X under X −→ S is
2D ≡ {a ∈ S | −1 (a) ⊆ D} = {a ∈ S |(∀x ∈ X)(x a → x ∈ D)}.
The weak (or existential) image of D under X −→ S is
3D ≡ {a ∈ S | −1 (a) G D} = {a ∈ S |(∃x ∈ X)(x a and x ∈ D)},
where −1 (a) G D denotes that the two sets intersect, i.e., that ∃x ∈ −1 (a) ∩D.
The right-hand sides in the above definitions show that the strong and weak images of a correspon-
dence are logically dual. The former gives the subset of S whose members can only be the image
of some point in D, whereas the latter gives the subset of S whose members can possibly be the
image of some point in D. By analogy with alethic modal logic, the strong image is thus written
as 2D (read “necessarily D”), while the weak image is written as 3D (read “possibly D”).2
In an entirely analogous way, the inverse correspondence S −1
−→ X generates both a strong and
weak inverse image of any U ⊆ S. In the literature, the strong image is called the restriction of U ,
while the weak image is the extent of U ; see for example Johnstone (1977) or Negri (2002).
Definition 2.3. The strong inverse image (or restriction) of U ⊆ S under X −→ S is
rest U ≡ {x ∈ X | (x) ⊆ U} = {x ∈ X |(∀a ∈ S)(x a → a ∈ U)}.
The weak inverse image (or extent) of U under X −→ S is
ext U ≡ {x ∈ X | (x) G U} = {x ∈ X |(∃a ∈ S)(x a and a ∈ U)}.
In addition to the operators 2, rest, 3, and ext, the topological interpretation of perceptions
requires two axioms. First, as indicated above, the agent must have some way, however coarse, of
perceiving anything that might be traded:2When more than one relation is being discussed, the symbols 2 and 3 are written with appropriate subscripts.
The notation here follows Sambin and Gebellato (1998) and Sambin (2001).
2.2 The Topological Interpretation of Perception 5
Axiom 2.1. For each x ∈ X, there is some a ∈ S such that x a.
Second, the agent’s perceptions must satisfy a consistency condition:
Axiom 2.2. For each a, a′ ∈ S, if there is some x ∈ X such that
x a and x a′,
then there is some a′′ ∈ S such that, for all x′ ∈ X,
x′ a′′ iff x′ a and x′ a′.
Thus, an agent who can perceive the same object in multiple ways must have a conception of objects
that can be possibly be perceived in these ways.
2.2 The Topological Interpretation of Perception
Under the stated axioms, it is possible to define the topology induced by perceptions. The process is
to compose the operations from X −→ S with the operations S −→ X, in order to define operators
from X −→ X, and then show that these operators are valid notions of interior and closure.
Intuitively, one can view the members of S as the names of open neighborhoods in the base of a
topology on X. That is, each a ∈ S is associated with
−1 (a) = {x ∈ X|x a},
which one can view as the neighborhood in X that a names. (Compare Valentini (2001).) Under
this interpretation, all of S becomes associated with the family
{{x ∈ X|x a}|a ∈ S}.
Definition 2.4 (Vickers (1988)). Let X be a topological space with base S. A subset D ⊆ X is
open iff every x ∈ D has a neighborhood N(x) ∈ S such that N(x) ⊆ D.
Associating S with neighborhoods in X modifies this definition as follows: D ⊆ X is open if and
only if, for every x ∈ D, there is some a ∈ S such that x a and −1 (a) ⊆ D.
6 2 PERCEPTION
Definition 2.5. The interior operator on X induced by is
int ≡ ext 2
Thus, for D ⊆ X, int D = ext2D.
Expanding this definition shows its justification: for D ⊆ X,
int D ≡ ext2D
= {x ∈ X|(∃a ∈ S)(x a and a ∈ 2D)}
= {x ∈ X|(∃a ∈ S)(x a and (∀x′ ∈ x)(x′ a → x′ ∈ D))}.
Noting that x a iff x ∈ −1 (a), the last expression above for int D says that it consists of those
points belonging to a neighborhood (named by some a ∈ S) whose points are all in D. That is, the
interior of an arbitrary D ⊆ X matches Definition 2.4.
Conversely, the following holds:
Proposition 2.1 (Sambin (2001)). For arbitrary D ⊆ X, int(int D) = int D.
Proof. This will be shown by showing 2ext2 = 2. The result then follows by composing both sides
on the left with ext.
Expanding definitions gives, for arbitrary D ⊆ X,
2ext2D = {a ∈ S|(∀x ∈ X)(x a → x ∈ ext2D)}
= {a ∈ S|(∀x ∈ X)(x a → (∃a′ ∈ S)(x a′ and (∀x′ ∈ X)(x′ a′ → x′ ∈ D)))}.
The last expression says that 2ext2D names neighborhoods whose points have a neighborhood a′
that is contained in D; thus, 2ext2D ⊆ 2D.
The reverse inclusion is immediate, by choosing a′ = a.
Given that the interior operator matches the classical definition and has the idempotent property
in Proposition 2.2, I make the following definition:
2.2 The Topological Interpretation of Perception 7
Definition 2.6. A subset D of X is open in the topology induced by iff
D = int D.
The standard definition of a closed set is one that contains all of its limit points. That is, the
closure of a set is the collection of points for which every open neighborhood intersects the set.
This would suggest that the closure of an arbitrary D ⊆ X should be defined as
{x ∈ X|(∀a ∈ S)(x a → (∃x′ ∈ X)(x′ a and x′ ∈ D))}.
Thus the natural definition of closure in this context is the logical dual of interior:
Definition 2.7. The closure operator on X induced by is
cl ≡ rest3
Thus, for D ⊆ X, cl D = rest3D.
An analogous argument to that in the Proposition 2.2 shows that 3ext3 = 3, and hence that
closure is idempotent. This justifies the following definition:
Definition 2.8. A subset D of X is closed in the topology induced by iff
D = cl D.
Expanding this definition for D ⊆ X gives
cl D ≡ rest3D
= {x ∈ X|(∀a ∈ S)(x a → a ∈ 3D)}
= {x ∈ X|(∀a ∈ S)(x a → (∃x′ ∈ X)(x′ a and x′ ∈ D))}
as desired. Intuitively, a real-world object is in the perceptual closure of D iff every way of perceiving
it is a way of perceiving something in D.
It can now be shown that an agent’s subjective perceptions induce a topology on X.
8 2 PERCEPTION
Lemma 2.1. In the perceptual topology, ∅ is clopen.
Proof. By definition,
int ∅ = ext2∅
= {x ∈ X|(∃a ∈ S)(x a and (∀x′ ∈ X)(x′ a → x′ ∈ ∅))} = ∅.
Thus, ∅ is open.
Analogously,
cl ∅ = rest3∅
= {x ∈ X|(∀a ∈ S)(x a → (∃x′ ∈ X)(x′ a and x′ ∈ ∅))} = ∅.
Thus, ∅ is closed.
Lemma 2.2. In the perceptual topology, X is clopen.
Proof. By definition,
int X = ext2X
= {x ∈ X|(∃a ∈ S)(x a and (∀x′ ∈ X)(x′ a → x′ ∈ X))
= {x ∈ X|(∃a ∈ S)x a}.
By Axiom 2.1, this is all of X, so X is open.
Analogously,
cl X = rest3X
= {x ∈ X|(∀a ∈ S)(x a → (∃x′ ∈ X)(x′ a and x′ ∈ X))} = X.
Thus, X is closed.
Lemma 2.3. The union of open sets in the perceptual topology is open.
Proof. By definition,
⋃α
int Dα = {x ∈ X|(∃α)(∃a ∈ S)(x a and (∀x′ ∈ X)(x′ a → x′ ∈ Dα))}
2.2 The Topological Interpretation of Perception 9
⊆ {x ∈ X|(∃a ∈ S)(x a and (∀x′ ∈ X)(x′ a → (∃α)(x′ ∈ Dα)))}
= int⋃α
int Dα.
To see the reverse inclusion, note that for any D ⊆ X, if x ∈ int D, then
(∃a ∈ S)(x a and (∀x′ ∈ S)(x′ a → x′ ∈ D)).
Picking x′ = x gives
x ∈ int D → x ∈ D,
i.e., int D ⊆ D. In particular,
int⋃α
int Dα ⊆⋃α
int Dα.
Combining these gives the result.
Lemma 2.4. The intersection of finitely many open sets in the perceptual topology is open.
Proof. It suffices to show that the intersection of two open sets is open, as the result then follows
by induction. For D,E ⊆ X,
int D⋂
int E = {x ∈ X|(∃a, b ∈ S)(x a and x b
and (∀x′ ∈ X)(x′ a → x′ ∈ D) and (∀x′′ ∈ X)(x′′ b → x′′ ∈ E))}.
By Axiom 2.2, if x a and x b, then there is some c ∈ S such that x c and
(∀x′ ∈ X)(x′ c → x′ a and x′ b),
which in turn implies,
(∀x′ ∈ X)(x′ c → x′ ∈ D⋂
E).
Thus, int D⋂
int E ⊆ int (D⋂
E).
Conversely, if x ∈ int (D⋂
E), then there is a neighborhood a ∈ S of x that is contained in D⋂
E,
which means that −1 (a) ⊆ D and −1 (a) ⊆ E. This says that x ∈ int D⋂
int E, which by the
arbitrariness of x implies int(D⋂
E) ⊆ int D⋂
int E.
Combining these shows that the finite intersection property holds.
10 2 PERCEPTION
The above lemmata are expressed in the following theorem:
Theorem 2.1 (Perceptual Topology). The open sets induced by perceptions, under Axioms 2.1
and 2.2, form a topology.
2.3 Remarks on Complementation
Nothing has been asserted about the negation of the relation . In particular, it has not been
stipulated whether (x, a) /∈ should be read as “x cannot be perceived as a,” or whether this
should merely indicate that the agent, faced with x, cannot say whether it is an instance of a.
Which reading one chooses determines the appropriate notion of complementation. Following
Bridges, Schuster, and Vıta (2002) and Vıta and Bridges (2003), I define the classical logical com-
plement of D ⊆ X as the members of X that do not belong to D:
¬D ≡ {x ∈ X|x 6∈ D}.
The classical logical complement of an open set is then those objects that are not perceived as
something in the set. Depending on how one reads absence from from , one might want a stronger
notion of complementation. Specifically, the apartness complement of D ⊆ X, written Dc, is the
set of objects that are always perceived as something outside of D:
Dc ≡ {x ∈ X|(∀a ∈ S)x a → a /∈ D}.
If the agent cannot say whether x is an instance of some a ∈ D, then x is in the classical logical
complement ¬D, but is not in the apartness complement Dc. For the two notions of complementa-
tion to coincide, the agent must be able to perceive exactly when a given object belongs to a given
set and when it does not. Since this reasoning applies in particular to every singleton {x} ⊂ X,
¬D = Dc if and only if the agent has perfect perceptions.
From the agent’s viewpoint, the apartness complement is the only notion of complementation that
can be useful. Note, however, that the apartness complement of an open set is not closed. This
feature, though somewhat strange to the uninitiated, is actually desirable: it reflects the fact that
perceptual limits prevent the agent from distinguishing boundaries.
11
Because closed sets are not defined as the complement of open sets (assuming one views the apart-
ness complement as the most useful notion), the topology induced by perceptions is intuitionistic
rather than classical. (See Brouwer (1907), Heyting (1956), Troelstra and van Dalen (1988), or
Dummett (2000) for details.) Thus existence proofs proofs must be constructive.
Note that this feature provides an alternative notion of unawareness to that in the literature (e.g.,
Fagin and Halpern (1988), Aumann and Brandenburger (1995), or Dekel, Lipman, and Rustichini
(1998b)). If there exists classically a solution to an agent’s problem, but no solution exists intu-
itionistically, the agent can be viewed as unaware of a solution.
3 Business Language
3.1 Basic Reporting Model and Topological Interpretation
For two agents to trade, they must be able to reach some sort of understanding about what they
are offering or requesting in exchange. Trades cannot be stated in terms of X, as agents do not
observe X directly, or even have mental conceptions of what is in X. Moreover, agents cannot offer
what is in the individual sets of conceptions Si, as these are private and subjective. I introduce a
shared language as a way around this difficulty. The terminology in the shared language is public,
so that the agents can use the language to try to reach some sort of consensual validation of what
objects are under discussion.
A business language is modeled as a set T , interpreted as shared terminology, and, for each i ∈ I,
a binary relation Ri between the agent’s private conceptions Si and T . The relation represents i’s
private semantics, i.e., how the agent may report a subjective conception in the business language.3
A report may be valid for more than one conception that the agent has in mind; in this case,
the conceptions that the agent can report in the same way are said to be within an immaterial
difference. Conversely, the same conception may have more than one way it can be reported.
3I do not discuss the development or evolution of the language, though there is a literature closely related to this
issue. See for example Ahn (2000).
12 3 BUSINESS LANGUAGE
Recalling the wine example from the introduction, if the language includes only the terms “full-
bodied” and “light-bodied,” then a wine taster may be able to report a medium-bodied wine using
either term, and conversely may have many wines in mind for which a given report is valid. This
is similar to the vague language in Lipman (2001), but because of the perceptual limits here, the
role of ambiguous language in this model is more beneficial than in Lipman’s.
For (a, t) ∈ Si×T , aRi t is read as, “a can be reported as t by agent i.” In parallel with the discussion
on perceptions, there are two correspondences associated with the agent’s reporting relation:
(∀a ∈ Si) Ri(a) ≡ {t ∈ T |aRi t}
and
(∀t ∈ Si) R−1i (t) ≡ {a ∈ Si|aRi t}.
The inverse correspondence gives the agent’s interpretation of what a report means. These two
correspondences generate the operations 3, ext, 2, and rest. Thus, for U ⊆ Si and W ⊆ T ,
3Ri(U) ≡ {t ∈ T |(∃a ∈ Si)(aRi t and a ∈ U)},
2Ri(U) ≡ {t ∈ T |(∀a ∈ Si)(aRi t → a ∈ U)},
extRi(W ) ≡ {a ∈ Si|(∃t ∈ T )(aRi t and t ∈ W )},
and
restRi(W ) ≡ {a ∈ Si|(∀t ∈ T )(aRi t → t ∈ W )}.
Reporting can be interpreted in terms of the topology (called the reporting topology) on an agent’s
conceptions. As in Theorem 2.1, the derivation requires two axioms:
Axiom 3.1. For each a ∈ Si, there is a t ∈ T such that aRi t.
Axiom 3.2. For each t, t′′ ∈ T , if there is an a ∈ Si such that
aRi t and aRi t′,
then there is some t′′ ∈ T such that, for all a′ ∈ Si,
a′ Ri t′′ iff a′ Ri t and a′ Ri t
′.
3.2 Communication 13
Axioms 3.1 is a non-degeneracy requirement. It says that there must be some way, however vague, of
reporting anything the agent may want to report. That is, the language must have some sufficiently
broad terms (“stuff,”for example) to cover anything. Axiom 3.2 requires consistency of the language:
if there are conceptions that can be reported more than one way, there must be a way to express
that there are multiple possible reports. Thus, if a wine could be called “full-bodied” or “light-
bodied,” then there must be a term such as “medium-bodied” for wines that could be reported in
both these ways.
The following definitions are analogous to those under perception:
Definition 3.1. The reporting interior operator is
intRi ≡ extRi2Ri .
The reporting closure operator is
clRi ≡ restRi3Ri .
A subset U of Si is open in the reporting topology if and only if U = intRi(U), and is closed in the
reporting topology if and only if U = clRi(U).
Theorem 3.1 (Reporting Topology). The open sets induced by the agent’s reporting relation, under
Axioms 3.1 and 3.2, form a topology.
Proof. Entirely analogous to the proof of Theorem 2.1.
3.2 Communication
The topological interpretations of perceptions and semantics make it possible to address how agents
with fundamentally different worldviews can nevertheless find a reliable way to trade. The idea
is that what one agent offers in the business language must be something that a trading partner
interprets as a faithful representation of what is actually traded, as the trading partner perceives it.
Because this subjective notion of a faithful representation differs from the accountant’s objective
notion (FASB 1980), I refer to the idea studied here as heterogeneous faithfulness.
14 3 BUSINESS LANGUAGE
As neither reports nor perceptions are in general unique, the business language cannot guarantee
that the same agent necessarily issues the same report when faced with the same object. The most
that can be required is that one agent reports what he or she sees in a way that a trading partner
would agree is a valid possible report.
Definition 3.2. Let i, j ∈ I be two agents, with conceptions Si, Sj , perception relations i, j ,
and reporting relations Ri, Rj for a set of common terminology T . The business language is het-
erogeneously faithful between i and j if and only if the following diagram commutes:
X i - Si
Sj
j
?
Rj
- T
Ri
?
That is, the business language is heterogeneously faithful if and only if
Ri◦ i= Rj◦ j .
If this holds for every i, j ∈ I, then the language is said to be heterogeneously faithful.
The following proposition shows that the direction of the definition could be reversed; that is, an
equivalent requirement is that two agents interpreting the same report have the same collection of
real-world objects in mind.
Proposition 3.1. Suppose a business language is heterogeneously faithful between two agents, i, j ∈
I. Then the interpretation of the reports is also heterogeneously faithful, i.e.,
−1i ◦R−1
i = −1j ◦R−1
j .
Proof. The heterogeneous faithfulness between i and j means that, given x ∈ X and t ∈ T , agent
i can report x as t iff agent j can do so also. For i to be able to report x as t, there must be some
a ∈ Si that is a way i can perceive x which i can report as t:
(∃a ∈ Si)(x i a and aRi t),
3.2 Communication 15
which also says that i can interpret t as x, i.e., that
(∃a ∈ Si)(a ∈ R−1i (t) and x ∈ −1
i (a)).
By an identical argument, if j can report x as t, then j can interpret t as x. Thus, heterogeneous
faithfulness between i and j means that the following square also commutes:
X � −1
i Si
Sj
−1j
6
�R−1
j
T
R−1i
6
Proposition 3.1 thus says that a language is useful for reporting entities if and only if it is useful
for end users. This feature depends on the fact that reporting and perception are defined by binary
relations and not necessarily by functions.
Theorems 2.1 and 3.1 show that 〈X, Si〉 and 〈Sj , T 〉 are topological spaces. Heterogeneous faith-
fulness thus requires each agent’s perceptions to induce a correspondence that carries collections of
open neighborhoods to collections of open neighborhoods, i.e., that takes open sets to open sets.
Thus, the condition here is related to lower hemi-continuity (Berge 1963):
Definition 3.3. A correspondence X −→ S is lower hemi-continuous if, for every U ⊆ S,
{x ∈ X| (x) G int U}
is open (where G denotes occupied intersection).
Theorem 3.2 (Heterogeneous Faithfulness and Lower Hemi-Continuity). If a business language
is heterogeneously faithful between two agents i, j ∈ I, then j’s perception correspondence j (·) is
a lower hemi-continuous correspondence between X, endowed with i’s perceptual topology, and Sj,
endowed with j’s reporting topology.
Conversely, suppose j (·) is a lower hemi-continuous correspondence between the topological spaces
〈X, Si〉 and 〈Sj , T 〉. Then there is in principle a reporting relation Ri(·) for i that makes the
language heterogeneously faithful.
16 3 BUSINESS LANGUAGE
Proof. By the proof of Proposition 2.2, 2 ext 2 = 2, and an analogous argument shows that
ext 2 ext = ext.
If U ⊆ Sj is the extent of some W ⊆ T , then int U = int(ext(W )) = ext 2 ext W = ext W = U .
Conversely, if U = int U , then automatically U is the extent of some W ⊆ T , namely 2U . Thus
U ⊆ Sj is open in the reporting topology if and only if it is the extent of some W ⊆ T , i.e., iff it is
the inverse image of some subset of T along Rj . A similar argument holds for an open D ⊆ X in
i’s perceptual topology.
The definition of lower hemi-continuity thus says that the inverse image of any W ⊆ T along
−1j ◦R−1
j is the extent of some subset U ′ ⊂ Si. So if W = extRiU′ for some U ′ ⊆ Si, then the
relation j is lower hemi-continuous. But this just says that the square below commutes:
X � −1
i Si
Sj
−1j
6
�R−1
j
T
R−1i
6
By Proposition 3.1, this is equivalent to heterogeneous faithfulness. Therefore, heterogeneous faith-
fulness implies lower hemi-continuity.
For the second part of the theorem, define the reporting relation for i by Ri ≡ Rj◦ j ◦ −1i .
The continuity of j means that Ri takes open neighborhoods in i’s perceptual topology to open
neighborhoods in j’s reporting topology, which is just the definition of heterogeneous faithfulness.
Remark. The phrase “in principle” in the second part of Theorem 3.2 reflects that the existence
is non-constructive. An agent using a language that is heterogeneously faithful can observe what
others say and can introspect; by so doing, the agent will fail to refute the claim that the language
has the desired property. Thus while the first portion of the theorem could not be established, it
could in practice at least withstand scrutiny.
The second part, however, requires more. To construct the desired reporting relation for agent i,
one would need access to Si and Sj (along with the various relations that are composed). This
17
would imply knowledge of others’ perceptions and of X, but the phenomena being studied is that
no one has such knowledge.
Accordingly, what the latter part of Theorem 3.2 establishes is that the non-existence of the desired
relation is contradictory. To one who is omniscient, this is equivalent to existence, but it is clear
that the relation used in the proof could not in practice be constructed. In light of the first
part of Theorem 3.2, one might view heterogeneous faithfulness as a form of continuity, classically
equivalent to lower hemi-continuity, but constructively stronger.4
There are in general many business languages that enable heterogeneously faithful reporting. Two
degenerate cases are the autarkic language, where T = ∅, and the universal language, where T is a
singleton (e.g, {“stuff”}). In both cases, the reporting relations are trivally faithful. In the autarkic
language, contracting is perfectly incomplete, and the only possible allocation is autarky. In the
universal language, any agents who agree to trade have no information on what they are bargaining
for, and provide no information on what they are offering, so that any economy is a grab bag. In
any other case, it is natural to ask what sort of allocations are attainable as equilibria. I turn to
this question in the next section.
4 Equilibrium and Welfare
4.1 Syntactical Requirements on the Language
Since all trade occurs in the shared language T , equilibrium must be stated in terms of T . For
market clearing to be a meaningful concept, T cannot be an arbitrary language, or even an arbitrary
heterogeneously faithful language. There is a syntactical requirement that an operation of addition
is defined on T , that T be closed under addition, that there is an additive identity (so that excess
demands can be said to sum to zero), and that every t ∈ T have an additive inverse (so that
it is possible to say how markets could clear from an arbitrary set of endowments). Moreover,
4Related ideas on continuity are discussed in Grandis (1997) and Gebellato and Sambin (2001). The notion of
heterogeneous faithfulness as a stability property of a language seems related to the strategic stability of Kohlberg
and Mertens (1986), as strategic stability is likewise a form of lower hemi-continuity.
18 4 EQUILIBRIUM AND WELFARE
the clearing of markets must be consistent with conservation of the physical and perceived flow of
goods. These requirements are summarized as follows:
Axiom 4.1. The shared language T is an Abelian (i.e., commutative) additive group, as are X and
the Si. The addition operation on T is common knowledge.
The perception and reporting relations cannot be assumed to be group homomorphisms. For
example, one cannot assume that (x+y) = (x) + (y): both x and y may differ imperceptibly
from 0, but their sum may be large enough to be distinct from 0. Similarly, a material quantity can
be split up into multiple immaterial quantities. It seems reasonable instead to require the following:
Axiom 4.2. For all i ∈ I, the correspondences induced by i and Ri are consistent with group
homomorphisms. That is, for x, y ∈ X, i (x + y) ⊆ i (x)+ i (y), and analogously for the Ri.
4.2 Optimality and Equilibrium
Notions of equilibrium and optimality depend on an agent’s preference relation, which for agent
i is assumed to be defined on Si. While there are philosophical reasons that one might want to
permit preferences to be incomplete (e.g., one might argue that conceptions that could be different
ways of perceiving the same object should not have a preference defined between them), a minimal
requirement would seem to be that strict preferences are irreflexive and transitive. Because the
absence of a strict preference might or might not be interpreted as a weak preference, I write a �i b
as a shorthand for ¬(a �i b). I then have the following notion of Pareto dominance:
Definition 4.1. Let (ai), (a′i) ∈∏
i∈I Si be perceived allocations. Allocation (ai) Pareto dominates
(a′i) iff, for every i ∈ I, a′i �i ai and, for some j ∈ I, aj �j a′j .
The notion of an agent’s lower contour set is likewise defined in terms of perceived consumption:
Definition 4.2. For i ∈ I and U ⊆ Si, the conceptions not better than U are
{a′i ∈ Si|(∃ai ∈ Si)(a′i �i ai and ai ∈ U)} ≡ ext�iU.
4.2 Optimality and Equilibrium 19
Given price functional Tp−→ R, agent i can afford to purchase t′ ∈ T if i’s perceived allocation can
be reported (under Ri) as a member of the following set:
I(p, t′) ≡ {t ∈ T |p · t′ ≤ p · t},
where, in keeping with convention, the value of t ∈ T under p is written p · t.
Definition 4.3. For agent i ∈ I with perceived endowment ai ∈ Si, some t′ ∈ T is budget feasible
for i if and only if
(∃t ∈ T )(ai Ri t and t ∈ I(p, t′)),
i.e., iff Ri(ai) G I(p, t′).
An optimal choice is now defined as follows:
Definition 4.4. For agent i ∈ I with perceived endowment ai ∈ Si, a choice of t′ ∈ T is optimal
given prices Tp−→ R iff
1. t′ is budget feasible, and
2. For every budget feasbile t′′ ∈ T , if there are b, c ∈ Si with b reportable as t′, c reportable as
t′′, and c �i b, then R−1i (t′′) G ext�i
R−1i (t′).
Thus, the agent must choose something that has at least one interpretation that is no worse than
any budget feasible alternative. Diagramatically, Definitions 4.3 and 4.4 say that optimality is a
commutative square:
SiRi - T
Si
ext�i
?
Ri
- T
I(p, ·)
?
This leads to the following result:
Lemma 4.1. A consumer’s budget problem has an optimal choice if and only if the agent’s prefer-
ences are lower hemi-continuous in the reporting topology.
20 4 EQUILIBRIUM AND WELFARE
Proof. Immediate consequence of Theorem 3.2 and Definition 4.4.
The definition of equilibrium is now straightforward: an equilibrium is budget feasible and optimal
for each agent, with markets clearing in the group operation defined in Axiom 4.1.
Definition 4.5. A competitive equilibrium is a price mapping Tp−→ R and a collection {(ti, t′i)i∈I}
of pairs in T ×T such that, if the initial perceived endowments are (ai ∈ Si)i∈I , then for each i ∈ I,
the following conditions hold:
Feasibility ai Ri ti and p · t′i ≤ p · t.
Optimality For feasible t′′i ∈ T , if (∃bi ∈ R−1i (t′i)) and (∃ci ∈ R−1
i (t′′i )) such that ci �i bi, then
there are b′i ∈ R−1i (t′i) and c′i ∈ R−1
i (t′′i ) such that c′i �i b′i.
Market Clearing∑
i∈I ti =∑
i∈I t′i, where the summations are in the group operation defined
on T by Axiom 4.1.
Since T is the base of a topology on Si, then if T is finite, every open cover of every subset of Si
(for every agent i ∈ I) necessarily has a finite subcover. In other words, T is finite, then every
set is compact. This would seem to be a powerful result for assuring the existence of equilibrium.
However, Axiom 4.1 restricts the cases where T can be finite. In particular, the following holds:
Proposition 4.1. If T is finite and addition is not isomorphic to addition mod n for some n ∈ Z++,
then T is a singleton.
Proof. See Lang (2000), pages 23–5. An intuitive argument is as follows: by Axiom 4.1, T is an
additive group. If t ∈ T and t 6= 0, then t + t ∈ T . If t + t = t, then t + (t + (−t)) = t + 0 = t,
violating the hypothesis. But then t+ t 6= t for every t 6= 0. If T is not a singleton, this means that
either addition must be isomorphic to addition mod n, which is ruled out by hypothesis, or T must
be infinite. Thus, either T = {0} with the only possible notion of addition, or T is infinite.
Nevertheless, it is always possible to coarsen T (even when T is infinite) enough in order to make
every set compact:
4.2 Optimality and Equilibrium 21
Definition 4.6. T is a Stone-compact space iff (∀W ⊆ T )(∃W0 ⊆ W ) such that
1. W0 is finite, and
2. ext W0 = ext W .
T is a Scott-compact space iff (∀W ⊆ T )(∀t ∈ T )
if ext {t} ⊆ ext W, then (∃t′ ∈ W )(ext {t} ⊆ ext {t′}).
Lemma 4.2. If T is Stone-compact, or if T is Scott-compact and I(p, ·) is finite-set valued for
every p, then a competitive equilibrium exists provided the language is heterogeneously faithful and
each agent’s preferences are lower hemi-continuous (in the sense of Lemma 4.1).
Proof. From heterogeneous faithfulness, preferences are consistent with continuous functions, and
in particular with continuous group homomorphisms (because of Axiom 4.2). Thus, each agent’s
optimization problem includes a continuous group homomorphism over a compact set, so every
agent has a budget feasible optimal choice, given a price functional, and in particular when prices
are market clearing.
Lemma 4.3 (Negri 1996). Ri and T always can be coarsened so that 〈Si, Ri, T 〉 is Stone-compact.
Proof. See Negri (1996), Proposition 2.8 and Theorem 2.10.
Theorem 4.1. A business language can always be coarsened so that competitive equilibrium exists.
Proof. Immediate corollary of Lemma 4.2 and Lemma 4.3.
Axioms 4.1 and 4.2 guarantee that equilibrium is consistent with conservation of the flow of goods;
this is the most that can be hoped for.
Proposition 4.2. Suppose that T is heterogeneously faithful and that individual preferences are
lower hemi-continuous in the sense of Lemma 4.1. Let ωi ∈ X be the initial real-world allocation
22 4 EQUILIBRIUM AND WELFARE
to agent i ∈ I, and let xi ∈ X be what agent i receives in equilibrium. If
∑i∈I
ωi =∑i∈I
xi,
then, for each i ∈ I, there are reports ti ∈ Ri ◦ i (ωi) and t′i ∈ Ri ◦ i (xi) such that markets
clear. Conversely, if markets clear in T , then there are (ωi) and (xi) satisfying the above equation,
i.e., such that the flow of goods is conserved.
Proof. From Axiom 4.1, the above expressions make sense. The correspondences from X −→ Si
and from Si −→ T contain continuous group homomorphisms because of Axiom 4.2, so their
composition contains a continuous group homomorphism.
4.3 Failure of the Welfare Theorems
The results above are promising, in that they show that competitive equilibrium exists for any
sufficiently coarsened business language. Unfortunately, much less can be said about the welfare
properties of equilibrium than in the standard neoclassical case. There can be competitive equilibria
that are Pareto dominated by feasible allocations, and there can be Pareto optimal allocations that
cannot be attained as a competitive equilibrium. Thus, neither welfare theorem holds. Because
of the topological interpretations, one might expect competitive equilibrium to be approximately
Pareto optimal, but even this is untrue. Thus, the welfare theorems do not seem robust to any
ambiguities in perception or language, or to any possible miscommunication.
In the non-heterogeneously faithful case, the failure of the welfare theorems is seen in the following:
Example 4.1. Let X = {x, y}, I = {1, 2}, S1 = {a1, b1}, S2 = {a2, b2}. Assume x 1 a1, x 2 a2,
y 1 b1, and y 2 b2. Thus each agent perceives X perfectly.
Let T have two terms, t1 and t2, and assume that agents 1 and 2 report differently, with a1 R1 t1,
b1 R1 t2, b2 R2 t1, and a2 R2 t2. That is, agent 1 uses t1 to mean x and t2 to mean y, and agent 2
reverses these. Since T is an additive group, assume T is given by all integer multiples of t1 and t2,
with addition defined componentwise.
4.3 Failure of the Welfare Theorems 23
Let preferences be given by
b1 �1 a1 a2 �2 b2,
and assume that initially agent 1 is endowed with x and agent 2 with y. Thus both agents can offer
t1 in trade and are willing to trade it for t2. If the price of t2 exceeds that of t1, then autarky is an
equilibrium. However, if both agents were to swap their endowments, they would both be better
off. That is, the allocation (b1, a2) is feasible and Pareto dominates the equilibrium (a1, b2).
In fact, the allocation (a1, b2) is Pareto optimal. However, if it is to emerge as equilibrium, then
both agents must offer to sell t1 at the equilibrium prices, and then purchase t1 at the equilibrium
prices. Since both agents will purchase t2 if it is affordable, in equilibrium the price of t2 must
exceed that of t1. If both agents sell their t1 and buy the other agent’s t1, then the Pareto optimum
is achieved. However, these equilibrium prices and reports are the same as those for which autarky
is an equilibrium. Therefore, no equilibrium prices assure trading from the Pareto dominated
endowments to the Pareto optimal allocation.
The above example depends on the lack of heterogeneous faithfulness. However, it is easy to see
that the welfare theorems fail even in the heterogeneously faithful case. The universal language
provides the simplest example:
Example 4.2. Let T = {0} be a singleton, with the only possible definition of addition. Then,
irrespective of X and the Si, every agent always issues the same report, and markets always clear.
Thus, the market is always in equilibrium. If there is any possible Pareto-dominated allocation,
it can arise as a result of the equilibrium, and nothing can assure a move to a Pareto-superior
allocation. That is, competitive equilibrium need not be optimal, and optimal allocations need not
be attainable as an equilibrium.
This example can be extended to any setting where an agent cannot distinguish in the reporting
language between conceptions over which the agent has a strict preference.
Examples 4.1 and 4.2 show how limits on the business language, in either the form of misun-
derstandings or ambiguities in interpretation, can make competitive equilibrium suboptimal. The
counter-examples do not depend on perceptual limits; indeed, Example 4.1 used perfect perceptions.
24 5 CONCLUDING REMARKS
It is straightforward to see that, when agents have perceptual limits, the language will necessarily
either have ambiguous semantics or fail to be heterogeneously faithful. Thus subjective perceptions
or subjective interpretations of the language conflict with the welfare theorems.
A natural way of ranking business languages is by their informativeness. Let T, T ′ be two sets
of shared terminologies, with reporting relations Ri, R′i for each agent i ∈ I. If, for each t ∈ T
and each i ∈ I, there is some t′ ∈ T ′ such that extR′i{t′} ⊂ extRi{t}, then T ′ is a (weakly) finer
information structure than T . This is closely related to Blackwell’s informativeness.
The results in this section show that ranking business languages by their informativeness is unlikely
to be a reasonable social preference. Once a language allows distinctions to become overly precise,
it ceases to be heterogeneously faithful; i.e., the agents can no longer use it and know what they
are trading. Thus, as in Dubra and Echenique ((2001) and (2004)), more detailed information can
be less desirable than coarser information.
5 Concluding Remarks
This paper studies the economic consequences of subjective information, by modeling how an agent
subjectively perceives real-world objects, and by modeling how such an agent can nevertheless
use a shared business language. The framework here is then used to study the interaction between
subjective perceptions and subjective semantics, and how these affect the way an economy functions.
The models of perceptions and of use of a language have useful topological interpretations. The
connection between what an agent perceives and what an agent reports is shown to be a form
of continuity in these topological spaces. Intuitively, the continuity condition says that an agent
reports a real-world object in a way that a trading partner, viewing the same object, could see
as justified. Because preferences are defined on the agents’ subjective worlds, the topological
interpretation of perceptions can be thought of as saying that agents’ preferences are defined on
open neighborhoods of consumption choices. Thus the model here might be viewed as an argument
in favor of Kreps’s (1979) definition of utilities on sets rather than on points, and with Jeffrey’s
(1983) definition of utility on propositions (hence on subsets). With some regularity conditions
25
on agents’ preferences, the topological interpretation is used again to provide conditions for the
existence of equilibrium.
Much less can be said about the welfare properties of competitive equilibria under subjective infor-
mation than in the usual case. Neither welfare theorem necessarily holds, even though preferences
are continuous and there are no externalities. The reason is that each agent does not see exactly
what is being traded, and cannot convey exactly what is desired. The invisible hand is therefore
not always able to allocate resources optimally, largely because the agents are unable to convey
precisely what they want it to do.
The notion of agents’ beliefs thus arises in a different sense from the usual probabilistic one. One
agent may observe another agent’s use of a language, and infer the distinctions that the other agent
is capable of making. Thus an extension of the current model to a dynamic model would enable
one to discuss beliefs about another agent’s subjective understanding of the world.
26 REFERENCES
References
Ahn, R. M. C. (2000): “Agents, Objects, and Events: A Computational Approach to Knowledge,
Observation, and Communication,” Ph.D. thesis, Technische Universitiet Eindhoven.
Allen, B., and J. Thisse (1992): “Price Equilibria in Pure Strategies for Homogeneous Oligopoly,”
Journal of Economics and Management Strategy, 1(1), 63–81.
Aumann, R. J., and A. Brandenburger (1995): “Epistemic Conditions for Nash Equilibrium,”
Econometrica, 63(5), 1161–80.
Avigard, J. (2004): “Forcing in Proof Theory,” Bulletin of Symbolic Logic, 10(3), 305–33.
Berge, C. (1963): Topological Spaces. Oliver and Boyd.
Bridges, D. S., P. Schuster, and L. S. Vıta (2002): “Apartness, Topology, and Uniformity:
a Constructive View,” Mathematical Logic Quarterly, 48(Supplement 1), 16–28.
Brouwer, L. E. J. (1907): “On the Foundations of Mathematics,” Ph.D. thesis, University of
Amsterdam, Translated by Arend Heyting and reprinted in L. E. J. Brouwer Collected Works I ,
North-Holland 1975.
Crawford, V. P., and J. Sobel (1982): “Strategic Information Transmission,” Econometrica,
50(6), 1431–51.
Dekel, E., B. L. Lipman, and A. Rustichini (1998a): “Recent Developments in Modeling
Unforeseen Contingencies,” European Economic Review, 42(3–5), 523–42.
(1998b): “Standard State-Space Models Preclude Unawareness,” Econometrica, 66(1),
159–73.
Dickhaut, J. W., and I. R. C. Eggleton (1975): “An Examination of the Process Underlying
Comparative Judgements of Numerical Stimuli,” Journal of Accounting Research, 13(4), 38–72.
Dubra, J., and F. Echenique (2001): “Monotone Preferences over Information,” Topics in The-
oretical Economics, 1(1), 1–15.
REFERENCES 27
(2004): “Information is not about Measurability,” Mathematical Social Sciences, 47(2),
177–85.
Dubra, J., F. Maccheroni, and E. A. Ok (2004): “Expected Utility Theory without the
Completeness Axiom,” Journal of Economic Theory, 115(4), 118–33.
Dummett, M. A. E. (2000): Elements of Intuitionism, no. 39 in Oxford Logic Guides. Oxford
University Press, second edn.
Fagin, R., and J. Y. Halpern (1988): “Belief, Awareness, and Limited Reasoning,” Artificial
Intelligence, 34, 39–76.
FASB (1980): “Qualitative Characteristics of Accounting Information,” Statement of Financial
Accounting Concepts No. 2, Financial Accounting Standards Board.
Gebellato, S., and G. Sambin (2001): “The Essence of Continuity (the Basic Picture, II),”
Preprint 27, University of Padua Department of Pure and Applied Mathematics.
Grandis, M. (1997): “Weak Subobjects and Weak Limits in Categories and Homotopy Categories,”
Cahiers de Topologie et Geometrie Differentielle Categoriques, 38, 301–26.
Heyting, A. (1956): Intuitionism: An Introduction. North-Holland.
Jeffrey, R. C. (1983): The Logic of Decision. University of Chicago Press, second edn.
Johnstone, P. T. (1977): Topos Theory. Academic Press.
Kanodia, C., R. Singh, and A. Spero (2005): “Imprecision in Accounting Measurement: Can
it be Value Enhancing?,” Journal of Accounting Research, 43(2), 487–519.
Knoblauch, V. (1998): “Order Isomorphism for Preferences with Intransitive Indifference,” Jour-
nal of Mathematical Economics, 30(4), 421–31.
Kohlberg, E., and J.-F. Mertens (1986): “On the Strategic Stability of Equilibria,” Economet-
rica, 54(5), 1003–38.
Kreps, D. M. (1979): “A Representation Theorem for ‘Preference for Flexibility’,” Econometrica,
47(3), 565–78.
28 REFERENCES
Lang, S. (2000): Algebra. Springer-Verlag, 3rd edn.
Lipman, B. L. (2001): “Why Is Language Vague?,” Discussion paper, Boston University.
Mas-Colell, A., M. D. Whinston, and J. R. Green (1995): Microeconomic Theory. Oxford
University Press.
Modica, S., and A. Rustichini (1999): “Unawareness and Partitional Information Structures,”
Games and Economic Behavior, 27(2), 265–98.
Negri, S. (1996): “Stone Bases, Alias the Constructive Content of Stone Representation,” in Logic
and Algebra, ed. by A. Ursini, and P. Agliano, pp. 617–36. Dekker.
(2002): “Continuous Domains as Formal Spaces,” Mathematical Structures in Computer
Science, 12, 19–52.
Rubinstein, A. (2000): Economics and Language. Cambridge University Press.
Sambin, G. (2001): “The Basic Picture, a Structure for Topology (the Basic Picture, I),”Discussion
paper, University of Padua Department of Pure and Applied Mathematics.
Sambin, G., and S. Gebellato (1998): “A Preview of the Basic Picture: A New Perspective on
Formal Topology,” in Types and Proofs for Programs, vol. 1657 of Lecture Notes in Computer
Science, pp. 194–207. Springer-Verlag.
Troelstra, A. S., and D. van Dalen (1988): Constructivism in Mathematics: An Introduction,
no. 121 and 123 in Studies in Logic and the Foundations of Mathematics. Elsevier, Two volumes.
Valentini, S. (2001): “Fixed Points of Continuous Functions Between Formal Spaces,” Discussion
paper, Department of Pure and Applied Mathematics, University of Padua.
Vickers, S. J. (1988): Topology via Logic, no. 5 in Cambridge Tracts in Theoretical Computer
Science. Cambridge University Press.
Vıta, L. S., and D. S. Bridges (2003): “A Constructive Theory of Point-Set Nearness,” Theo-
retical Computer Science, 503(1–3), 473–89.